News

04.03.2020 - Die Präsentationsfolien sind nun im Moodle verfügbar.

14.01.2020 - Die Seminarveranstaltung findet in der nächsten Woche am Dienstag, den 21.01. und am Mittwoch, den 22.01. in Raum OH14 / E04 statt. Der Start ist jeweils 8:00 Uhr. Weitere Informationen zum Ablauf stehen unter Programm.

02.12.2019 - Reviewers assigned

31.07.2019 - Topics assigned

09.07.2019 - Homepage uploaded

Next Deadlines

01.03.2020 - Abgabe der finalen Ausarbeitung über HotCRP (incl. Vortragsfolien)

21.01.2020 / 22.01.2020 - Seminarvorträge

23.12.2019 - Abgabe der zwei Reviews über HotCRP

29.11.2019 - Abgabe der Ausarbeitung über HotCRP

23.12.2019 - Abgabe der drei Reviews

01.03.2020 - Abgabe der finalen Version der Ausarbeitung und der Präsentationsfolien über HotCRP

Description

Hardware accelerators are widely used in modern systems. Dedicated coprocessors such as GPUs, Xeon Phis, or FPGAs provide speed-ups over CPUs. Vice-versa modern CPUs integrate accelerators such as SIMD units to process multiple elements per instruction.

Leveraging such hardware accelerators in a beneficial way is a relevant topic for research and industry. Not only the implementation of highly parallel operations, but also the integration with existing systems remains a challenge. The latter due to communication bottlenecks, cross-platform code bases, and off-loading decision making.

This seminar is about hardware acceleration for modern in-memory query processing techniques. In the seminar, we will discuss a range of topics that addresses various hardware accelerators in the context of column-at-a-time processing, vectorized query processing, and query compilation.

Programm

Das Blockseminar findet am Dienstag, den 21.02. und am Mittwoch, den 22.02 mit Beginn jeweils um 8:00 Uhr in Raum OH14 / E04 statt. Im Folgenden steht ein vorläufiges Programm für den Ablauf der beiden Tage. Weitere Details zum Ablauf besprechen wir am Dienstag Morgen zu Beginn der Veranstaltung.

Session 1 -- Di 21.02. 8-10 Uhr

A1. Gregg et al. Where Is the Data? Why You Cannot Debate CPU vs. GPU Performance Without the Answer. IEEE ISPASS 2011
Timo Gojowczyk

A2. Wu et al. Kernel Weaver: Automatically Fusing Database Primitives for Efficient GPU Computation. MICRO 2012
Philipp Koppenstein

A4. Yuan et al. The Yin and Yang of processing data warehousing queries on GPU devices. VLDB 2013
Philipp-Jan Honysz

Session 2 -- Di 21.02. 10-12 Uhr

B1. Karnagel et al. Optimizing GPU-Accelerated Group-By and Aggregation. ADMS 2015
Richard Stewing

B2. Jiang et al. Efficient SIMD and MIMD parallelization of hash-based aggregation by conflict mitigation. ICS 2017
Beka Chkopoia

C1. Heimel et al. Hardware-Oblivious Parallelism for In-Memory Column-Stores. VLDB 2013
Dustin Chabrowski

Session 3 -- Mi 22.02. 8-10 Uhr

C2. Breß et al. Generating custom code for efficient query execution on heterogeneous processors. VLDB J. 2018
Robert Delhougne

D1. Karnagel et al. Adaptive Work Placement for Query Processing on Heterogeneous Computing Resources. VLDB 2017
Marius Gunnemann

D2. Breß et al. Robust Query Processing in Co-Processor-accelerated Databases. SIGMOD 2016
Jannis Muehlemeyer

Session 4 -- Mi 22.02. 10-12 Uhr

E1. Merrill et al. Single-Pass Parallel Prefix Scan with Decoupled Look-Back. NVIDIA Tech. Report 2016
David Kaester

F1. Teubner et al. Data Processing on FPGAs (Chapters 1, 2, and 3). Synthesis Lectures on Data Management 2013
Pajtim Thaqi

F2. Woods et al. Parallel Computation of Skyline Queries. FCCM 2013
Nils Killich

Session 5 -- Mi 22.02. ca. 13:30-15 Uhr

F3. Teubner et al. Skeleton Automata for FPGAs: Reconfiguring without Reconstructing. SIGMOD 2012
Fabian Dillkötter

F4. Ziener et al. FPGA-based dynamically reconfigurable SQL query processing. TRETS 2016
Shimon Wonsak

Organization

Preliminary Meeting

A preliminary meeting took place on Friday, July 12th 2019 at 9:15 AM in OH14 / room 302.

Information Document

We provide an information document for the seminar. The document describes the course schedule and gives you guidelines for working on the presentation and report. You can download the document here.

Report

The report should follow the ACM Proceedings Templates (LaTeX or Word) with a maximum of 6 pages. Each student reviews two reports from other students.

Talk

The talks are scheduled for 25 to 30 minutes. Each talk is followed by a discussion on the topic and on the presentation.

Topics

A. Coprocessor Communication

• A1. Gregg et al. Where Is the Data? Why You Cannot Debate CPU vs. GPU Performance Without the Answer. IEEE ISPASS 2011 – pdf
  Timo Gojowczyk

• A2. Wu et al. Kernel Weaver: Automatically Fusing Database Primitives for Efficient GPU Computation. MICRO 2012 – pdf
  Philipp Koppenstein

• A3. Funke et al. Pipelined Query Processing in Coprocessor Environments. SIGMOD 2018 – pdf 
  Marcel Ebbrecht

• A4. Yuan et al. The Yin and Yang of processing data warehousing queries on GPU devices. VLDB 2013 – pdf
  Philipp-Jan Honysz

B. Relational Operator Implementations

• B1. Karnagel et al. Optimizing GPU-Accelerated Group-By and Aggregation. ADMS 2015 – pdf
  Richard Stewing

• B2. Jiang et al. Efficient SIMD and MIMD parallelization of hash-based aggregation by conflict mitigation. ICS 2017 – pdf
  Beka Chkopoia

C. Cross-Platform Processing

• C1. Heimel et al. Hardware-Oblivious Parallelism for In-Memory Column-Stores. VLDB 2013 – pdf
  Dustin Chabrowski

• C2. Breß et al. Generating custom code for efficient query execution on heterogeneous processors. VLDB J. 2018 – pdf
  Robert Delhougne

D. Operator Placement

• D1. Karnagel et al. Adaptive Work Placement for Query Processing on Heterogeneous Computing Resources. VLDB 2017 – pdf
  Marius Gunnemann

• D2. Breß et al. Robust Query Processing in Co-Processor-accelerated Databases. SIGMOD 2016 – pdf
  Jannis Muehlemeyer

E. Parallel Primitives

• E1. Merrill et al. Single-Pass Parallel Prefix Scan with Decoupled Look-Back. NVIDIA Tech. Report 2016 – pdf
  David Kaester

• E2. Egielski et al. Massive Atomics for Massive Parallelism on GPUs. ISMM 2014 – pdf
  Julia Moldenhauer

F. FPGAs

• F1. Teubner et al. Data Processing on FPGAs (Chapters 1, 2, and 3). Synthesis Lectures on Data Management 2013 – pdf
  Pajtim Thaqi

• F2. Woods et al. Parallel Computation of Skyline Queries. FCCM 2013 – pdf
  Nils Killich

• F3. Teubner et al. Skeleton Automata for FPGAs: Reconfiguring without Reconstructing. SIGMOD 2012 – pdf
  Fabian Dillkötter

• F4. Ziener et al. FPGA-based dynamically reconfigurable SQL query processing. TRETS 2016 – pdf (Uninetz)